Memory retrieval strategies in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

How does a key-value memory network retrieve information?

In a key-value memory network, what is the main step used to retrieve the correct value from memory?

A Matching the input query with keys to find the most relevant value

B Randomly selecting a value from memory without using the input

C Using the last stored value regardless of the query

D Sorting all values and returning the first one

💡 Hint

Think about how the network finds the right information based on the input.

The key-value memory network compares the input query to stored keys to find the closest match, then returns the associated value.

❓ Predict Output

intermediate

Output of memory attention weights calculation

What is the output of the following code that calculates attention weights over memory keys?

Agentic AI

import numpy as np
query = np.array([1, 0])
keys = np.array([[1, 0], [0, 1], [1, 1]])
scores = keys @ query
weights = np.exp(scores) / np.sum(np.exp(scores))
print(weights)

A [1. 0. 0.]

B [0.33333333 0.33333333 0.33333333]

C [0.5 0.5 0.0]

D [0.42214066 0.1553624 0.42214066]

💡 Hint

Check how the dot product scores are computed and then converted to probabilities.

The scores are dot products of keys and query: [1, 0, 1]. Applying softmax gives the weights shown.

❓ Hyperparameter

advanced

Choosing the right memory size for retrieval accuracy

When increasing the size of an external memory in a retrieval model, which effect is most likely to occur if the memory size is too large without proper regularization?

A The model may retrieve irrelevant information, reducing accuracy

B The model will always improve accuracy with larger memory

C The model will ignore the memory and rely only on input features

D The model will crash due to memory overflow errors

💡 Hint

Think about how too much information can confuse retrieval.

Too large memory without constraints can cause the model to retrieve noisy or irrelevant data, hurting accuracy.

🔧 Debug

advanced

Identify the error in memory retrieval code snippet

What error will the following code raise when trying to retrieve a value from memory?

Agentic AI

memory = {'apple': 'fruit', 'carrot': 'vegetable'}
query = 'banana'
value = memory[query]
print(value)

A TypeError

B KeyError

C ValueError

D No error, prints 'fruit'

💡 Hint

What happens if you try to access a dictionary key that does not exist?

Accessing a missing key in a dictionary raises a KeyError in Python.

❓ Model Choice

expert

Best memory retrieval model for long-term context in dialogue systems

Which model architecture is best suited for retrieving relevant long-term context in a dialogue system with very large memory?

A Feedforward neural network with fixed-size input

B Simple RNN without external memory

C Differentiable Neural Computer (DNC) with content-based addressing

D Convolutional Neural Network (CNN) for image classification

💡 Hint

Consider models designed to read and write to external memory with flexible addressing.

DNCs use content-based addressing to read/write from large external memory, making them ideal for long-term context retrieval.